MiniMax teased M3 Sparse Attention: 9.7x prefilling, 15.6x decoding at 1Mtwitter.com9 points·rebekkamikkoa··0 commentsOpen articleSaveView on HN